update docs

Author: Chester Curme

README.md

@@ -6,7 +6,7 @@ Ghost allows you to chat with (1) a LLM that is fine-tuned on your text messages
 ## Configuration
 Ghost requires two configuration files to tailor your chat responses:
 1. Describe your background, writing style, or mannerisms by modifying [prompt_prefix.md](ghost/settings/prompt_prefix.md), or setting the `PROMPT_PREFIX_PATH=/path/to/prompt_prefix.md` environment variable.
-1. Provide structured information such as your SMS-enabled Twilio phone number, chatbot name, and aliases in [contacts.json](ghost/settings/contacts.json). This configuration file also houses an array of facts that is indexed and retrieved over when generating responses. You can also point to this configuration file by setting the `CONTACTS_PATH=/path/to/contacts.json` environment variable.
+1. Provide structured information such as your SMS-enabled Twilio phone number, chatbot name, and aliases in [contacts.json](ghost/settings/contacts.json). If using the Langchain agent, this configuration file also houses an array of facts that is indexed and retrieved over when generating responses. You can also point to this configuration file by setting the `CONTACTS_PATH=/path/to/contacts.json` environment variable.
 
 The [contacts.json](ghost/settings/contacts.json) file specifies other important information such as allowed senders for incoming messages, and information about the senders that helps Ghost tailor its responses. See [settings/README.md](ghost/settings/README.md) for more detail.
 

ghost/fine_tuning/README.md

@@ -18,3 +18,6 @@ This workflow is set up in a [Jupyter notebook](notebooks/sample_records.ipynb),
 * `path_to_output_train_data`: path to which to write a .jsonl file of training data for OpenAI. Ghost will format the conversations into prompts and completions using the information in [prompt_prefix.md](../settings/prompt_prefix.md) and [contacts.json](../settings/contacts.json).
 
 ## Fine-tuning a model
+The train data output by the above sampling step is of the format expected by OpenAI's fine tuning endpoints. We provide a [sample notebook](notebooks/fine_tune.ipynb) to illustrate how .jsonl files sampled from different contacts can be concatenated and uploaded to the [fine-tunes/create](https://platform.openai.com/docs/api-reference/fine-tunes/create) endpoint. N.B. OpenAI's [documentation](https://platform.openai.com/docs/guides/fine-tuning) includes helpful hints on fine tuning for the purpose of creating a conversational chat application, such as using fewer epochs and a smaller learning rate.
+
+Once your model is created, updating the `"fine_tuned_model_name"` field in [contacts.json](../settings/contacts.json) to the model's `fine_tuned_model` field configures the application to route messages through the fine tuned model.